Horizontal pod splitting scheduling for cost-effectiveness in cloud computing

Optimizing service latency while controlling resource costs is a critical challenge in cloud-native microservice deployments. This paper proposes Horizontal Pod Splitting (HPS), a novel scheduling method that jointly optimizes the number of pod replicas and resource quotas under fixed total resources to minimize service latency. We construct a double-exponential service latency model that characterizes the relationships among total resources, unit resource concurrency, pod replicas, and service latency. Based on this model, we theoretically derive the optimal number of pod replicas and schedulability condition. Experimental results on a Kubernetes cluster demonstrate that HPS reduces service latency by 6–12% compared to HPA and by 7–22% compared with Libra under equivalent resource conditions, while achieving higher cloud resource utilization and reducing deployment costs for service providers.